Meta's New AI Model Lets Users Drop Other Instagram Users Into Their Photos
Muse Image, the first release from Meta's Superintelligence Labs, can pull a real account's public photos into an AI-generated image with a simple @ mention.

Meta is rolling out a new AI image generator called Muse Image, and its most striking feature is one that raises immediate questions about consent and identity online: type an @ mention of another Instagram account into a prompt, and the model will pull that person's public photos into the image it generates.
The company announced Muse Image on Tuesday as the first model built by Superintelligence Labs, the AI division Meta assembled last year. It's now powering image tools inside the Meta AI app, Instagram, and WhatsApp, with Facebook and Messenger integration coming soon.
According to Meta, tagging a username lets the AI use that person's public photos to help build a visual. Meta says people can control how others reuse their content for AI purposes, though the announcement doesn't spell out exactly what those controls look like or how granular they are. The basic mechanic is straightforward enough: instead of describing a person in text and hoping the model approximates them, a user can just reference an actual Instagram account and let Muse Image work from real, public images of that person.
Part of a bigger shift away from Llama
Muse Image belongs to what Meta calls its Muse family of models, which is gradually replacing the Llama lineup that had been the backbone of the company's AI efforts. Alexandr Wang, the entrepreneur Meta hired last year to run Superintelligence Labs, described Muse Image on Threads as "agentic," meaning it doesn't just generate an image from a prompt in one shot. Instead, it pairs with Meta's Muse Spark large language model to, in Wang's words, "reason through your prompt, search the web, and plan before it generates."
That's a meaningfully different approach than a typical text-to-image model, which maps a prompt directly to pixels. Muse Image is designed to interpret intent, gather relevant information, and plan out a result before producing it. Wang also teased an upcoming Muse Video model, claiming it will be "competitive on prompt adherence, visual fidelity, temporal consistency" — though Meta hasn't given a launch date for it.
What the model can actually do
Beyond the @ mention feature, Muse Image comes with a set of practical tools. Users can transform existing images using suggested prompts, generate designs for things like invitations and postcards, and redesign a room based on a photo pulled from Facebook Marketplace or elsewhere on the web — useful, say, for visualizing how a piece of furniture might look in your space. There's also a draw-on-photo feature that lets people sketch directly on top of an image to indicate the changes they want, then share the result to their feed, story, or a chat.
The model is also the engine behind 30 new AI effects launching in Instagram Stories, starting in the US before expanding to other countries and further into Meta's app ecosystem.
Why the tagging feature stands out
Most mainstream AI image tools have steered away from letting users generate images of identifiable real people without their direct involvement, precisely because of the potential for misuse — harassment, impersonation, or non-consensual imagery. Meta's approach threads that needle by tying the feature to the existing social graph: you're not describing a stranger, you're tagging an actual account, and Meta says users retain some say over how their content gets used this way.
Still, the feature is a notable bet. It leans into Instagram's core asset — a massive, interconnected network of public profiles and photos — as training and generation material for AI images, in a way few other platforms can replicate. Whether the safeguards Meta describes are enough to prevent unwanted or embarrassing uses of someone's likeness will likely become clearer once the feature sees wider, real-world use.
For now, Muse Image marks a clear signal of where Meta's AI ambitions are headed: agentic systems that reason and plan rather than just generate, and tools built to exploit the social data Meta already has at massive scale. The Muse Video model Wang teased suggests this is just the opening move.


